# Rubric Generation on OpenAI HealthBench HARD Source: [Rubric Generation on OpenAI HealthBench HARD]() Collection index: [llms.txt]() Data updated: 2026-03-24T13:25:38.030Z Evaluation of model performance on generating rubrics using the OpenAI HealthBench HARD dataset, measured via standard NLP metrics (BLEU, ROUGE) and LLM-based judging. A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | Rubric Generation | | Dataset (including any stated split) | OpenAI HealthBench HARD | | Metric | BLEU | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 6 of 6 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [BLEU](): Higher is better; unit: PERCENT. - [ROUGE-1](): Higher is better; unit: PERCENT. - [ROUGE-2](): Higher is better; unit: PERCENT. - [ROUGE-L](): Higher is better; unit: PERCENT. - [LLM Judge Score](): Higher is better; unit: PERCENT. ## Recorded results | Method | BLEU | Result date | Report paper | | --- | --- | --- | --- | | SFT | 4.2 | Not recorded | 69c21a700636e5bf4afc4716 | | RubricRAG | 3.9 | Not recorded | 69c21a700636e5bf4afc4716 | | RubricRAG | 3 | Not recorded | 69c21a700636e5bf4afc4716 | | Few-Shot | 2.9 | Not recorded | 69c21a700636e5bf4afc4716 | | GRPO | 2.7 | Not recorded | 69c21a700636e5bf4afc4716 | | Zero-Shot | 1.5 | Not recorded | 69c21a700636e5bf4afc4716 | ### SFT BLEU: 4.2. Result ID: 69c290c647f34ce924aad8a8. [Reporting paper]() Recorded evaluation configuration: {"Mode":"SFT"}. This record does not assert independent reproduction. ### RubricRAG BLEU: 3.9. Result ID: 69c290c647f34ce924aad8aa. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Thinking":"nothink"}. This record does not assert independent reproduction. ### RubricRAG BLEU: 3. Result ID: 69c290c647f34ce924aad8a9. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Thinking":"think"}. This record does not assert independent reproduction. ### Few-Shot BLEU: 2.9. Result ID: 69c290c647f34ce924aad8a6. [Reporting paper]() Recorded evaluation configuration: {"Mode":"Few-Shot"}. This record does not assert independent reproduction. ### GRPO BLEU: 2.7. Result ID: 69c290c647f34ce924aad8a7. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Mode":"GRPO"}. This record does not assert independent reproduction. ### Zero-Shot BLEU: 1.5. Result ID: 69c290c647f34ce924aad8a5. [Reporting paper]() Recorded evaluation configuration: {"Mode":"Zero-Shot"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.